linux-kernel - Re: dm stripe: add DAX support

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <1468362104.8908.43.camel@hpe.com>
Date:	Tue, 12 Jul 2016 22:22:05 +0000
From:	"Kani, Toshimitsu" <toshi.kani@....com>
To:	"snitzer@...hat.com" <snitzer@...hat.com>
CC:	"linux-raid@...r.kernel.org" <linux-raid@...r.kernel.org>,
	"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
	"dan.j.williams@...el.com" <dan.j.williams@...el.com>,
	"dm-devel@...hat.com" <dm-devel@...hat.com>,
	"ross.zwisler@...ux.intel.com" <ross.zwisler@...ux.intel.com>,
	"linux-nvdimm@...1.01.org" <linux-nvdimm@...1.01.org>,
	"agk@...hat.com" <agk@...hat.com>
Subject: Re: dm stripe: add DAX support

On Fri, 2016-06-24 at 14:29 -0400, Mike Snitzer wrote:
> 
> BTW, if in your testing you could evaluate/quantify any extra overhead
> from DM that'd be useful to share.  It could be there are bottlenecks
> that need to be fixed, etc.

Here are some results from fio benchmark.  The test is single-threaded and is
bound to one CPU.

 DAX  LVM   IOPS   NOTE
 ---------------------------------------
  Y    N    790K
  Y    Y    754K   5% overhead with LVM
  N    N    567K
  N    Y    457K   20% overhead with LVM

 DAX: Y: mount -o dax,noatime, N: mount -o noatime
 LVM: Y: dm-linear on pmem0 device, N: pmem0 device
 fio: bs=4k, size=2G, direct=1, rw=randread, numjobs=1

Among the 5% overhead with DAX/LVM, the new DM direct_access interfaces
account for less than 0.5%.

 dm_blk_direct_access 0.28%
 linear_direct_access 0.17%

The average latency increases slightly from 0.93us to 0.95us.  I think most of
the overhead comes from the submit_bio() path, which is used only for
accessing metadata with DAX.  I believe this is due to cloning bio for each
request in DM.  There is 12% more L2 miss in total.

Without DAX, 20% overhead is observed with LVM.  Average latency increases
from 1.39us to 1.82us.  Without DAX, bio is cloned for both data and metadata.

Thanks,
-Toshi